Skip to content
sundar@portfolio: ~/portfolio
← homeBook a call
sundar@portfolio:~$whoami

Sundar Shahi Thakuri · Full-stack Product Engineer

$ whoami

Sundar Shahi Thakuri · Full-stack Product Engineer

$ since

2018 · 6+ years shipping production SaaS

$ roles

enroute → ktmlabs → krispcall → covosys

$ certs

✓ Google Cloud · Professional Cloud Architect

$ open-source

207 repos · 20 collaborations

$ status

available for full-stack & cloud work

6+
years_shipping
207
github_repos
8
products_live
20
collaborations
sundar@portfolio:~$origin
where_it_started

Before the first device, I wrote code on paper

I started coding in 8th grade, long before I owned a computer. With no laptop in reach, I wrote programs out by hand on paper and traced them line by line in my head.

When all I had was a phone, that's where I typed and ran them, building from whatever device was within reach instead of waiting for the right one.

That instinct never left: reason the whole thing through, then build it from whatever the moment hands you.

then

No computer. Code on paper, traced in my head, run on a borrowed phone.

now

I automate the engineering I can. Agentic dev and tight build-and-verify loops do the real work, so my attention goes to the hard calls instead of the keystrokes.

AI agentsloop engineeringCI/CD
sundar@portfolio:~$arc

# arc: front-end to full-stack, one role at a time

  1. Three of these four roles were front-end by title, and I've kept it that way. The breadth came from what I took on inside them, and from building full products alongside them. I don't relabel the past; I show the arc.

  2. 2018 – 2020

    Frontend Engineer, React

    EnRoute Nepal

    • Translated designs and wireframes into high-quality, accessible React code.
    • Authored internal front-end guidelines and best practices for the team.
    • Mentored junior engineers in JavaScript and React.

    stack-reach: frontend (foundations)

  3. 2020

    Software Engineer, Front-end

    KTMLabs

    • Optimized front-end systems for performance, reliability, and maintainability.

    stack-reach: frontend

  4. 2020 – 2023

    Software Development Engineer, Front-end

    KrispCall

    • Built a real-time React/TypeScript call interface over a WebRTC voice stack, driven by GraphQL subscriptions instead of polling.
    • Shipped reusable component libraries used across multiple products.
    • Worked cross-functionally with backend, product, UX, and QA.

    stack-reach: frontend (real-time, deep)

    See it shipped

  5. milestone

    Google Cloud · Professional Cloud Architect

    Certification earned · cloud → certified

  6. 2023 – present

    current

    Senior Software Engineer

    CovoSys

    • Designs and delivers robust, client-focused solutions across multiple products, from interface down to infrastructure.
    • Leads feature delivery end to end, ensuring timely, high-quality releases.

    stack-reach: frontend · backend · cloud

  7. shipped

    AI enters the stack through a product, not a title

    SparkTales

    • Built the Next.js app and the whole LLM generation pipeline: structured prompting, guardrails, and a generation UX tuned to the latency users actually felt.

    stack-reach: ai · backend · cloud

    Read the case study
sundar@portfolio:~$stack

# stack: shown rather than claimed (drag to orbit)

frontend

ReactNext.jsTypeScriptJavaScriptGraphQLTailwindFigma

backend

Node.jsPythonGraphQL

cloud

Google Cloud (certified)AWSKubernetesTerraformDocker

tooling

GitHub · 207 reposCI/CD

GCP-certified; cloud and AI are where I've grown most.

sundar@portfolio:~$work

# proof_of_range: The range, in shipped products.

Three shipped products that show the breadth isn't theoretical. Each one is the receipt for a layer above.

proof of range

proof of range

Textquill · On-device OCR + offline grammar checker (Chrome extension)

Textquill

On-device OCR + offline grammar checker (Chrome extension)

A free Chrome extension that pulls editable text out of any image, screenshot, PDF, or screen region — then proofreads, translates, and rewrites it, all without a server. V2 turned an OCR tool into a full local text pipeline: four WebAssembly models (OCR, grammar, translation, Chrome's built-in AI) running on the user's own hardware, 38 languages, and an always-on grammar checker that never uploads a draft. Lives at textquill.app.

JavaScriptChrome MV3WebAssemblyonnxruntime-webChrome built-in AI
Read case study38 OCR languages

proof of range

Self-hosted speech-to-text · Speech recognition engine for a low-resource language
SPEECH AI

Self-hosted speech-to-text

Speech recognition engine for a low-resource language

Voice messages were reaching an AI assistant as an empty placeholder, because no affordable speech API handled the language well. I fine-tuned an open-source recognition model on a free GPU and shipped it as a self-hosted API on hardware that was already paid for: word error rate 0.808 to 0.379, and the wrong-alphabet failure eliminated.

PythonPyTorchONNX RuntimeFastAPIDocker
Read case study0.808 → 0.379 word error rate, same held-out set

proves: real-time frontend + cloud (GCP)

proves: full-stack + AI (Next.js · LLM · Node · GCP)

proves: full-stack product build (Next.js · Node)

proof of range

Webhook flood controls · Ingest hardening for a multi-tenant CRM (client engagement)
PLATFORM

Webhook flood controls

Ingest hardening for a multi-tenant CRM (client engagement)

A partner platform's replay loop delivered 487,195 events into a CRM's webhook endpoint in eleven hours and pinned the box at 99% CPU. I traced it a week after the fact through a session-manager shell, found that all three existing protections were structurally incapable of catching it, and built the ingest budget that closes the class of failure rather than the instance.

Ruby on RailsPostgreSQLGoodJobAWS SSMDocker
Read case study487,195 events in 11 hours

the_full_stack_read

Where the arc lands: full-stack, earned.

The titles say front-end three times, and I'll stand by that. But across these roles I moved from interfaces into APIs, into cloud architecture (Google Cloud Professional Cloud Architect), and into AI features shipped inside live products. Today I take a product from interface to infrastructure, with the AI in between.

frontend

proven by KrispCall
ReactNext.jsTypeScriptGraphQLTailwind

backend

proven by SparkTales & Keelspot
Node.jsPythonGraphQL

cloud

✓ GCP runs KrispCall & SparkTales
GCP (certified)AWSKubernetesTerraformDocker

ai

✓ proven by SparkTales
LLMsAgentsVertex AI

Cloud and AI are real strengths within the range (you can verify them in the case studies), not the whole story.

All case studies →
sundar@portfolio:~$say

# what_others_say: What people I've worked with say.

“His commitment to growth (his own and the team's) made for a genuinely motivating place to work. Collaborative, technically strong, and a natural leader.”
Kiran DuwalJava Software Developerenroute_nepal · worked together
“Intelligent and endlessly curious, with a real knack for learning. Dedicated, dependable, and someone you could always count on to get things done.”
Aashish ManandharReact Developer · Software Engineerenroute_nepal · 2019–2020 · I mentored him
“Exceptional and deeply dedicated. He doesn't just finish the work, he finds a better way to do it. A great team lead and a great colleague.”
Manish RanaSoftware Engineer, ReactJS/NextJSteam lead
“Highly productive and multi-skilled, with broad knowledge, and always willing to help on others' projects too.”
Raghav GuptaDC Consultant, Deloitte USI · ex-Wipro · ex-Newgencolleague
sundar@portfolio:~$contact

# work_with_me

You've seen the path. Now tell me about yours.

A 20-minute call. Tell me the problem, and I'll tell you how I'd approach it, from the interface down to the infrastructure.

Book a callrésumé:PDFDOCXLinkedInGitHub